Dastgàh recognition in Iranian music: different features and optimized parameters

Abstract

In this paper we report on the results of utilizing computational analysis to determine the dastgàh, the mode of music in the Iranian classical art music, using spectrogram and chroma features. We contrast the effectiveness of classifying music using the Manhattan distance and Gaussian Mixture Models (GMM). For our database of Iranian instrumental music played on a santur, using spectrogram and chroma features , we achieved accuracy rates of 90.11% and 80.2% when using Manhattan distance respectively. When using GMM with chroma, the accuracy rate was 89.0%. The effects of altering key parameters were also investigated, varying the amount of the training data and silence, as well as high frequency suppression on the results. The results from this phase of experimentation indicated that a 24 equal temperament was the best tone resolution. While experiments focused on dastgàh, with only minor adjustments the described techniques are applicable to traditional Persian, Kurdish, Turkish, Arabic and Greek music, and therefore suitable to use as a basis for a musicological tool that provides a broader form of cross-cultural audio search.